“If the prompts for Claude tell us anything, it’s that without human guidance and hand-holding, these models are frighteningly blank slates.”
Anthropic posted the July 12 system prompts for Claude 3 Opus, 3.5 Sonnet and 3 Haiku and promises to log every change. The prompts tell Claude to act “very smart and intellectually curious,” to be “completely face blind,” and never to open with “certainly.” The only way to get GPT-4o’s prompt is a prompt injection attack. The personality is a page of stage directions, and now at least one vendor shows you the script.